Firms in International Trade (with an Application to Spain) Pol Antràs Puchal (Harvard University) Rubén Segura-Cayuela (Banco de España) with Diego Rodríguez Rodríguez (U. (U Complutense) December 2010 SERIES Invited Lecture 1 Overview O i off R Recent Developments 2 Neoclassical Trade Theory Firms are treated as a black box (supply side = production set) Often assume constant Oft t t returns t to t scale, l so firm fi size i is i indeterminate General equilibrium only pins down the size of the sector or industry to which the firm belongs Very powerful theory, but of limited use when studying firm-level issues in international trade December 2010 SERIES Invited Lecture 3 New Trade Theory Introduced increasing returns, imperfect competition and product differentiation Thi helped This h l d resolve l the h indeterminacy i d i off firm fi size i With product differentiation, firms face downward sloping demand curves and there exists an optimal firm size Free entry (and general equilibrium) then pins down industry size and also the number of firms within an industry New Trade Theoryy rationalized two-wayy trade flows in similar products across countries December 2010 SERIES Invited Lecture 4 Some Problematic Predictions As insightful as New Trade Theory is, it delivers some counterfactual predictions B Because all ll firms fi within i hi a sector are treatedd symmetrically, either all firms within an industry export or none does (and they always do with CES) Trade liberalization generally affects all firms within an industry symmetrically (and when it doesn doesn’tt, there is no way to predict these asymmetries) These predictions are problematic because they do not provide a good description of reality December 2010 SERIES Invited Lecture 5 “Evidence” “There is no good reason to believe that the assumptions of the Dixit-Stiglitz model – a continuum of goods that enter symmetrically into demand, with the same cost functions, and with the elasticity of substitution between any two goods both constant and the same for any pair you choose – are remotely true in reality.” December 2010 SERIES Invited Lecture 6 “Evidence” “There is no good reason to believe that the assumptions of the Dixit-Stiglitz model – a continuum of goods that enter symmetrically into demand, with the same cost functions, and with the elasticity of substitution between any two goods both constant and the same for any pair you choose – are remotely true in reality.” Paul Krugman, g Nobel Lecture 2008 December 2010 SERIES Invited Lecture 7 Real Evidence: Heterogeneity in Data Standard deviation of log sales Productivity: Standard deviation of log value added per worker for U.S. plants: Overall: 0.75 Within 4-digit sectors (450 sectors): 0.66 December 2010 SERIES Invited Lecture 8 Exporters are in the Minority December 2010 SERIES Invited Lecture 9 Exporters in the U.S. (4-digit) December 2010 SERIES Invited Lecture 10 Exporters Are Different than Non-Exporters December 2010 SERIES Invited Lecture 11 Exporters Are Different than Non-Exporters December 2010 SERIES Invited Lecture 12 Interpreting the Evidence An obvious question at this point is: Do differences in performance generate selection into exporting, or does exporting generate differences in performance? Not straightforward to tease out empirically: One can look at the timing of productivity changes and exporting (does exporting lag productivity improvements or vice versa) But notice that firms can select into exporting because they anticipate that their productivity is in an upward trend (Costantini and Melitz, Melitz 2008) December 2010 SERIES Invited Lecture 13 Empirical Tests Strong evidence for self-selection of more productive firms into exporting Colombia, C l bi Mexico, M i andd Morocco: M Clerides, Cl id Lach, L h andd Tybout (1998, QJE) U.S.: Bernard and Jensen ((1999,, JIE)) Taiwan: Aw, Chen, and Roberts (2001, JDE) Mixed evidence for “learning-by-exporting” g y p g Some evidence in growing, developing countries (India, Slovenia) – see De Loecker (2007, JIE) December 2010 SERIES Invited Lecture 14 Caveats Exogenous causality from either export status or productivity is suspect N evidence New id shows h that h fi firms make k joint j i t decisions d ii concerning both export status and technology choice: Verhoogen (2009 (2009, QJE): quality upgrade and exports in Mexico Bustos ((2010, AER): ) new exporters p in Argentina g spend p more on technological upgrades Lileeva and Trefler (2010, QJE): similar for Canada December 2010 SERIES Invited Lecture 15 Effects of Trade Liberalization There is now mounting evidence that trade liberalization induces important reallocation effects E Exporters expand, d non-exporters contract, andd this hi raises industry productivity Chil trade Chile: t d liberalization lib li ti in i 1979-85 1979 85 lledd to t 19% productivity increase (of which 2/3 is explained by reallocation effects) Similarly for Canadian firms after U.S.-Canada free trade agreement (Trefler, 2004) December 2010 SERIES Invited Lecture 16 Plant Death and Exporting Bernard and Jensen (ReStat, 2007): Unconditionally, they find that in the U.S. export status is associated with 12 12.6% 6% reduction in probability of death (this is large, overall probability is 27%). Is it jjust that low-productivity p y firms are more likelyy to die and these tend to be non-exporters? No. Conditional on a full set of industryy and firm controls (productivity, size, capital-labor ratio, ...), export status is still associated with a significant 5-6% reduction d i in i probability b bili off death d h December 2010 SERIES Invited Lecture 17 Why Do We See These Effects? “It is a capital mistake to theorize before one has all the evidence. Insensibly one begins to twist facts to suit theories theories, instead of theories to suit facts facts” December 2010 SERIES Invited Lecture 19 Why Do We See These Effects? “It is a capital mistake to theorize before one has all the evidence. Insensibly one begins to twist facts to suit theories theories, instead of theories to suit facts facts” Sherlock Holmes (1891) December 2010 SERIES Invited Lecture 19 Towards Succesful Theories 1. 2. Evidence suggests that successful theoretical frameworks for studying firms and the decision to export should include two features: Within sectoral heterogeneity in size and productivity A feature that leads only the most productive firms to engage in foreign trade: fixed cost of exporting (Melitz, 2003) variable markups p (BEJK, ( , 2002,, Melitz and Ottaviano,, 2008) December 2010 SERIES Invited Lecture 20 Melitz (2003) Eachh firm E fi produces d its it own “variety” “ i t ” off a goodd (CES) Developing this variety entails a sunk entry cost Following entry entry, firms observe their productivity Prior to entry, only distribution of potential productivity levels is known (common for all firms) Firms also face a fixed overhead production cost Exporting p g involves both a standard “per-unit” p trade cost as well as a fixed export cost An entering firm decides whether to produce (or exit) and then whether to export (or only serve home market) December 2010 SERIES Invited Lecture 21 Melitz (2003): Equilibrium December 2010 SERIES Invited Lecture 22 Melitz (2003): Equilibrium December 2010 SERIES Invited Lecture 23 Melitz (2003): Reallocation Effects December 2010 SERIES Invited Lecture 24 Trade Liberalization Forces least productive firms to exit (competitive pressure) R ll Re-allocates market k shares h towards d more productive d i firms Resulting in higher average productivity Welfare gains: combination of higher average productivity and ambiguous effect of product variety But quantitatively, not clear there are additional gains from trade given certain observable variables (Arkolakis et al., 2010) December 2010 SERIES Invited Lecture 25 Aggregate Implications Melitz (2003) model is successful in accounting for several micro facts in the data M More iimportantly, l this hi micro-founded i f d d model d l off industry equilibrium has generated important new insights for the aggregate response of exports to shocks The increase in aggregate productivity is just one example, but many others have been highlighted December 2010 SERIES Invited Lecture 26 Chaney (2007) Develops multi-sector, multi-country Melitz model with Pareto distribution of productivity Sh Shows that h model d l predicts di modified difi d gravity i equation i Standard gravity Gravity with heterogeneous firms December 2010 SERIES Invited Lecture 27 Chaney (2007) A larger denotes a smaller right-tail of the Pareto distribution Thus, the elasticity of trade flows to changes in trade frictions is a function of characteristics of the size distribution of firms This elasticity is lower in sectors/countries with high dispersion in firm size (Key: extensive margin) December 2010 SERIES Invited Lecture 28 Helpman, Melitz & Rubinstein (08) Also demonstrate that firm-level models of exporting can have important implications for aggregate bilateral trade flows In particular, they can easily explain the large number of zeros observed in these flows Similar to Chaney (2007) but they develop an econometric approach for estimating trade flows Is being widely used in several applications December 2010 SERIES Invited Lecture 29 Helpman, Melitz & Rubinstein (08) Trade in both directions Trade in one direction only No trade 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% December 2010 SERIES Invited Lecture 19 96 19 94 19 92 19 90 19 88 19 86 19 84 19 82 19 80 19 78 19 76 19 74 19 72 19 70 0% 30 Di Giovanni & Levchenko (2010) They note that if the slope of the Pareto distribution is close to 1, then aggregate exports closely track those of the largest firms in an economy Implication: if productivity or costs evolve differently for large and small firms, firms the evolution of large firms may be particularly relevant Simple p pprice indices might g not capture p this correctlyy They estimate the slope for several countries and find an estimate veryy close to 1 in most countries December 2010 SERIES Invited Lecture 31 Policy Implications Interesting question: conditional on some average level of productivity, is there a socially optimal size distribution of firms? Still very underdeveloped area – results seem very model specific December 2010 SERIES Invited Lecture 32 An Application to Spain 33 Road Map Can these new approaches shed light on the behavior of aggregate Spanish exports during the 2000s? W will We ill argue that h this hi behavior b h i is i puzzling li from f the h point of view of homogenous firms models A more micro-founded i f d d investigation i ti ti provides id new insights that bring us closer to explaining the puzzle Still we will see that the stylized models developed Still, so far are too basic to account for all the patterns we observe December 2010 SERIES Invited Lecture 34 Puzzle: Competitiveness Falls… December 2010 SERIES Invited Lecture 35 Puzzle: Competitiveness Falls… December 2010 SERIES Invited Lecture 36 The Puzzle: but Market Share is Flat December 2010 SERIES Invited Lecture 37 The Puzzle: but Market Share is Flat December 2010 SERIES Invited Lecture 38 Market Share is Flat in Most Sectors December 2010 SERIES Invited Lecture 39 Puzzle: Homogenous Firm Models Relative exports should be decreasing in relative export prices (and, in turn, in relative costs) SPAIN SPAIN X SPAIN DSPAIN P C X X X OTHER OTHER X OTHER D OTHER P C X X X With CES preferences, this looks SPAIN SPAIN X OTHER X X ψ OTHER X ψ C C SPAIN X OTHER X OTHER / ψ where ψSPAIN is a relative demand shifter X X December 2010 SERIES Invited Lecture 40 Explanations for Puzzle 1. Relative Demand Shifts a. b. 2. Relative Supply Shifts a. b. 3 3. Increase in relative quality of Spanish exports Hi h iincome growth Higher th in i predominant d i t Spanish S i h importers i t Effects Eff t off heterogeneity h t it Markup adjustment (though relative prices are going up) Capital Flows, Flows FDI, FDI and Current Account Dynamics We will focus on 2.a., 2 a but will say a word about the others December 2010 SERIES Invited Lecture 41 Heterogeneity H i in i Spanish S ih Exporters 42 Spanish Exporters Also Perform Better l X it D expit Ind ln I dit Year Y it Si Sizeit it Exporting Premia Source: ESEE Dependent Variable (X): All firms Output Employment Capital per worker l k Capital per hour Results: Wage per worker Output per worker Output per hour Output per hour N. of observations 0.479 (33.74) 0.086 (14.9) 0,253 (18,2) 0,255 (17,9) 0.106 (18.91) 0.385 (30.53) 0.386 (30.61) 17,740 Firms with 200 or fewer workers fewer workers 1.233 (57.47) 0.692 (44.66) 0,496 (32,7) 0,503 (32,5) 0.190 (32.37) 0.536 (43.41) 0.539 0.539 (43.66) 12,589 Firms with more than 200 workers than 200 0.222 (4.26) 0.195 (6.00) 0,122 (3,6) 0,118 (3,4) 0.035 (2.26) 0.022 (0.56) 0.022 (0.57) 5,151 December 2010 SERIES Invited Lecture 43 But There is Substantial Heterogeneity Source: ESEE December 2010 SERIES Invited Lecture 44 In 2000s, Large Exporters Grew More Source: ESEE December 2010 SERIES Invited Lecture 45 In 2000s, Large Exporters Grew More Source: ESEE December 2010 SERIES Invited Lecture 46 In 2000s, Large Exporters Grew More Source: ESEE December 2010 SERIES Invited Lecture 47 Why This Differential Growth? Source: ESEE December 2010 SERIES Invited Lecture 48 Why This Differential Growth? Source: ESEE December 2010 SERIES Invited Lecture 49 Why Might This Affect Aggregate Exports? December 2010 SERIES Invited Lecture 50 Is This Sufficient to Explain the Puzzle? No! We are not saying anything quantitative at this point More importantly, what matters is relative competitiveness What if large firms in other countries are seeing their competitiveness rise by even more? F this For hi we needd firm-level fi l l data d from f other h countries i Homogenized dataset available in a few weeks Next: preliminary evidence from OECD and Amadeus December 2010 SERIES Invited Lecture 51 Relative Competitiveness December 2010 SERIES Invited Lecture 52 Change in Relative Competitiveness Source: Amadeus December 2010 SERIES Invited Lecture 53 Recap We observe heterogeneous performance of Spanish exporters Gi Given thickness hi k off right-tail, i h il these h changes h in i relative competitiveness have potential to explain aggregate export behavior But: why are price indices not capturing this? Weighting does not seem to appropriately take into account for intraindustry heterogeneity December 2010 SERIES Invited Lecture 54 S Some N Notes on Al Alternative i Explanations 55 A Relative Demand Shock? Sources: WB, ECB, WTO, and MEH December 2010 SERIES Invited Lecture 56 A Relative Demand Shock? Sources: WB, ECB, WTO, and MEH December 2010 SERIES Invited Lecture 57 A Markup Adjustment? Sources: WB, ECB, WTO, and MEH December 2010 SERIES Invited Lecture 58 A Markup Adjustment? Sources: WB, ECB, WTO, and MEH December 2010 SERIES Invited Lecture 59 A Markup Adjustment? Source: ESEE December 2010 SERIES Invited Lecture 60 A Markup Adjustment? Dependent variable: Growth of Sale Prices Coefficient t-statistic Growth of Intermediate Input Prices 0.3168 39.81 L Large Firms Fi 0 1514 0.1514 1 90 1.90 S Source: ESEE December 2010 SERIES Invited Lecture 61 “Quality” Improvements? Sources: WB, ECB, WTO, and MEH December 2010 SERIES Invited Lecture 62 Is It Explained by Inward FDI? Source: ESEE December 2010 SERIES Invited Lecture 63 Is It Explained by Inward FDI? S Source: ESEE December 2010 SERIES Invited Lecture 64 Is It Explained by Outward FDI? Source: ESEE December 2010 SERIES Invited Lecture 65 Loose Ends 66 The Non-Monotonicity Around 2003 Sources: WTO and BdE December 2010 SERIES Invited Lecture 67 Recent Crisis Ch Changes iin L Labor b C Costs and dP Productivity d i i ((percentages)) Firm Size (no. of employees) 200 or less More than 200 2008 2009 2008 2009 Total Labor Costs A Average T Total t l Employment E l t 1,9 -5,3 53 -8,7 -9,4 94 3,3 -2,6 26 -7,7 -8,9 89 Real Ouput of Goods and Services -5,0 -18,0 -8,4 -17,6 Real Value Added -0,1 -6,1 -8,0 -9,6 Labor cost per worker Productivity (Output based) 7,6 0,3 0,8 -9,5 6,0 -6,0 1,3 -9,6 Productivity (Value added based) 5,5 3,6 -5,5 5,5 -0,8 0,8 Source: ESEE December 2010 SERIES Invited Lecture 68 (In)Conclusions 69 (In)Conclusions Puzzling behavior of aggregate Spanish exports in light of apparent loss of competitiveness M d l with Models i h firm fi heterogeneity h i appear promising i i in i (partly) explaining this puzzle Loss of competitiveness not homogenous across firms: large firms less affected Other simple explanations appear at odds with features of the data Future work: use of new homogenized g dataset and quantification December 2010 SERIES Invited Lecture 70 (In)Conclusions “Sorprenderse, extrañarse, es comenzar a entender” (“To be surprised, to wonder, is to begin to understand.” ) José Ortega y Gasset December 2010 SERIES Invited Lecture 71